Research Archives - HPC Montenegro

24.10.2026.24.10.2026.

MSc Thesis Defence: Explainable AI for Type 2 Diabetes Risk Assessment

On October 21, 2026, MSc candidate Sanja Lazić successfully defended her thesis entitled “Interpretable Machine Learning and Deep Learning Models for Type 2 Diabetes Risk Prediction Based on Behavioural and Clinical Indicators” within the Artificial Intelligence Master’s programme at the University of Donja Gorica, Faculty for Information Systems and Technologies, under the mentorship of Prof. Dr. Tomo Popović.

The research and experimental work were partially supported by NCC Montenegro through HPC infrastructure and expertise within the EuroCC3 project, with additional research support from the AI-AGE project. Through its involvement in academic research, mentoring, and competence development, NCC Montenegro contributes to strengthening the application of advanced AI and HPC technologies in healthcare and other scientific domains.

Ms Sanja Lazić during her MSc thesis defence

The thesis investigates the application of machine learning, deep learning, and explainable AI techniques to type 2 diabetes risk assessment, focusing on predictive performance, reliability, and interpretability. The research also demonstrates how AI models can support preventive healthcare through accessible, questionnaire-based risk assessment without requiring laboratory measurements.

ABSTRACT – This thesis presents the development and systematic evaluation of machine learning and deep learning models for type 2 diabetes risk assessment using behavioural, clinical, and demographic indicators. The study analysed 253,680 records and compared five models: Logistic Regression, Random Forest, XGBoost, LightGBM, and a deep neural network. Model performance was evaluated in terms of discrimination, probability calibration, and interpretability. While the deep neural network achieved the highest ROC-AUC (0.7728), LightGBM demonstrated the most balanced overall performance. Explainable AI methods, including TreeSHAP, Integrated Gradients, and DeepLIFT SHAP, were applied to identify factors contributing to model predictions. A prototype web application was developed to demonstrate how estimated risks and their explanations could be presented to users.

The results demonstrate the potential of explainable AI for supporting non-invasive risk assessment and preventive screening, while highlighting the importance of model reliability, transparency, and further validation before clinical application.

13.10.2026.24.10.2026.

EuroHPC-Supported Research Advances AI-Based Lecture Transcription for Montenegrin

Results from the Exams 4.0 Proof of Concept project, supported by the Innovation Fund of Montenegro and EuroHPC computing resources, were presented at the international SAMYRAD 2026 conference in Seville, Spain.

The paper “Whisper-Based Lecture Transcription for Montenegrin Higher Education with a Practical Evaluation” was presented at the Smart Algorithms & Models: Your Reach to Advanced Dialogues (SAMYRAD 2026) conference, held on 5–6 October 2026 in Seville, Spain.

Authored by Stevan Čakić, Tomo Popović, Srđan Krčo, and Igor Čulafić, the paper investigates the use of OpenAI Whisper models for automatic transcription of university lectures in Montenegrin, a language not directly supported by Whisper.

The study evaluated three Whisper model variants and three related South Slavic language settings using 30 university lectures covering nearly 38 hours of audio. The experimental work involved generating 4,860 transcripts, with large-scale processing performed on the LUMI EuroHPC supercomputer in Finland.

The results identified Whisper’s turbo variant as the most suitable among the tested models in terms of transcription consistency, evaluated text quality, and processing speed. The research also provided practical recommendations for processing lecture recordings in Montenegrin.

These findings have already been incorporated into a new lecture-transcription module developed within the Exams 4.0 platform, enabling lecturers to transform audio and video recordings into text that can support AI-assisted preparation of educational materials and examination questions.

This R&D effort was conducted as part of the Exams 4.0 Proof of Concept project, implemented by DigitalSmart in collaboration with the University of Donja Gorica and funded by the Innovation Fund of Montenegro.

HPC Montenegro (NCC Montenegro) supported the research team in securing access to EuroHPC computing resources through the EuroHPC JU Development Access programme (project ID: EHPC-DEV-2024D12-018). The computational experiments were performed on LUMI, hosted by CSC in Finland and the LUMI consortium.

This work demonstrates how access to European HPC infrastructure can support applied AI research, foster innovation in higher education, and contribute to the development of digital tools for smaller and underrepresented languages.

11.10.2026.11.10.2026.

AI Research on Premium Domain Name Recommendation Published in IEEE Access

A new scientific paper titled “Premium Domain Name Recommendation via Semantic Retrieval and Reranking of Fixed Aftermarket Inventories” has been published in the international journal IEEE Access. The research was conducted by Milutin Pavićević, Tomo Popović, Srđan Krčo, and Miodrag Vujković from the University of Donja Gorica (UDG) and DunavNET.

The study addresses the challenge of recommending relevant premium domain names from large collections of domains already available for resale. The researchers developed and evaluated an approach combining large language models (LLMs), semantic embeddings, vector search using FAISS, and different reranking strategies. Experiments were conducted on 30,000 domain names sampled from a collection of nearly 700,000 premium domains, with evaluations involving 22 industry experts and 180 students. The results demonstrated the potential of semantic retrieval to improve domain name recommendations while balancing relevance, diversity, and computational cost.

The research also has clear potential for industrial application. Alicorn, a Montenegro-based software company specializing in solutions for the domain name industry, is already developing AI-powered domain name recommendation technologies, including its Name Sentinel platform. The methods investigated in this study could contribute to further improving such solutions, particularly through more effective semantic search and recommendation of premium domains.

The research represents another example of research and innovation supported by European initiatives and connected with the activities of HPC Montenegro (NCC Montenegro), contributing to the development of AI expertise and stronger links between academia and industry.

Read the scientific paper: IEEE Access – DOI: 10.1109/ACCESS.2026.3740779

25.09.2026.11.10.2026.

AI-AGE Research Demonstrates the Potential of Retinal AI for Colorectal Cancer Detection

A new scientific paper developed within the AI-AGE research project, implemented by the University of Donja Gorica (UDG) and the University of Montenegro, has been published in the international journal Current Medical Imaging. The study, titled “Deep Learning Discrimination of Colorectal Cancer using Retinal Fundus Photographs: UK Biobank Case-Control Study,” explores the potential of artificial intelligence (AI) to identify retinal imaging patterns associated with colorectal cancer.

Using 1,600 retinal fundus images from the UK Biobank dataset, the researchers developed a deep learning model based on EfficientNet-B3, achieving an AUROC of 0.945 on an independent participant-level test set. The findings suggest that retinal images may contain valuable information about systemic diseases beyond the eye, opening possibilities for future non-invasive screening and risk assessment approaches. However, further clinical validation is required before such methods can be considered for practical healthcare applications.

The research represents an important achievement for the AI-AGE project and demonstrates the growing potential of advanced AI methods in biomedical research in Montenegro. The project benefits from synergies with HPC Montenegro (NCC Montenegro) and the EuroCC initiatives, which support the development of national competencies in High-Performance Computing (HPC), data analytics, and artificial intelligence. This collaboration contributes to strengthening the research infrastructure and expertise needed for computationally demanding medical AI applications.

The AI-AGE project is implemented by the Faculty for Information Systems and Technologies at UDG and the Faculty of Medicine at the University of Montenegro, with funding from the Ministry of Education, Science and Innovation of Montenegro.

Read more:

27.07.2026.27.07.2026.

BSc Thesis: BabyFlow.AI – A Mobile Assistant for Early Parenthood Support

Marija Gardašević successfully defended her BSc thesis, “Application of Artificial Intelligence in the Development of a Mobile Assistant for Early Parenthood Support,” at the Faculty for Information Systems and Technologies, University of Donja Gorica. Supervised by Prof. Tomo Popović, the thesis resulted in BabyFlow.AI, a functional Flutter-based mobile prototype for recording and reviewing a baby’s daily routines, including sleep, feeding, mood, meals and basic health-related information. The prototype may be considered an early proof of concept for a future digital parenting-support product, subject to further user testing, integration of more advanced analytics and validation in cooperation with relevant professionals.

ABSTRACT – The thesis presents the development of BabyFlow.AI, a functional mobile application prototype designed to help parents organize and monitor a baby’s daily routine. The application combines modules for sleep, feeding, pumping, mood, meals, calming-noise sessions, vaccination records, growth tracking and relevant health-information resources, while storing user data locally on the device. Basic analytical insights are generated from user entries, and a separate demonstrative Python machine-learning model illustrates a simple data-processing and classification workflow. BabyFlow.AI is not intended for medical, clinical or diagnostic use, but represents a practical proof of concept that could support the future development of privacy-conscious digital services for parents and childcare providers.

27.07.2026.27.07.2026.

BSc Thesis: AI and HPC for Cybersecurity of Connected Vehicles

Anđela Vuković successfully defended her BSc thesis, “Application of AI and HPC Technologies in the Cybersecurity of Connected Vehicles,” at the Faculty for Information Systems and Technologies, University of Donja Gorica. Supervised by Prof. Tomo Popović, with the support of Arnad Lekić, the thesis combines an analysis of automotive cybersecurity with a practical AI model for detecting anomalies in CAN communication. The developed demonstrator can be regarded as an early proof of concept for potential industrial applications, subject to further testing with real vehicle data and deployment-oriented validation.

ABSTRACT – This thesis examines the application of Artificial Intelligence (AI) and High-Performance Computing (HPC) in the cybersecurity of connected vehicles. In addition to analysing contemporary cybersecurity risks, communication technologies and protection mechanisms, the thesis presents a practical implementation of an autoencoder model for detecting anomalous CAN messages. The experimental model demonstrated the potential of AI-based anomaly detection, while HPC is considered an important enabler for processing larger datasets, training more advanced models and simulating cyberattack scenarios. Although developed using simulated data, the implementation provides an initial proof of concept that could be extended and validated for use by automotive manufacturers, cybersecurity providers, testing laboratories and fleet operators.

30.06.2026.10.07.2026.

MSc Thesis Defence: Quantization of Edge AI Models in IoT Systems

On June 29, 2026, an MSc thesis entitled “Quantization of Edge AI Models in IoT Systems” by Mr. Zarko Perunicic was successfully defended within the Artificial Intelligence Master’s programme at the University of Donja Gorica. Through its participation in the programme, mentoring activities, and support for practical research in AI, HPC, and IoT, NCC Montenegro contributes to developing advanced competencies in the efficient deployment of artificial intelligence models on resource-constrained devices. The thesis addresses an important Edge AI challenge by evaluating model quantization strategies for computer vision applications in IoT environments.

ABSTRACT – Edge AI systems in Internet of Things (IoT) environments require artificial intelligence models that are sufficiently small, fast, and reliable to operate on resource-constrained devices. This thesis examines how quantization, as a model optimization method, affects the performance of a computer vision model in the task of grape leaf disease classification. MobileNetV2 was used as the reference model, and its optimized variants were then prepared in the TensorFlow Lite environment using FP16 and INT8 quantization modes, including dynamic INT8 quantization, full INT8 quantization based on a representative dataset, and an INT8 variant obtained through quantization-aware training (QAT) on an additional, more challenging dataset. The experiments were conducted on cleaned and restructured subsets, following quality control of publicly available datasets and the removal of redundant and visually equivalent samples. Under controlled conditions, latency, execution stability, peak RAM usage, model size, and accuracy were analyzed.

On the more controlled dataset, full post-training INT8 quantization achieved the most favorable balance among efficiency, stability, and model size while preserving accuracy, whereas dynamic INT8 quantization, despite reducing model size, can measurably slow down model execution. On the more challenging field dataset, this pattern changed partially: although full INT8 quantization remained the fastest variant, the INT8 model obtained through QAT provided the most favorable overall balance between accuracy, model size, and latency. The results show that the effect of quantization depends not only on numerical precision, but also on data characteristics, the calibration procedure, and the compatibility of the model with the execution environment. It is therefore concluded that the choice of quantization strategy should be empirically validated for a specific application scenario rather than assumed in advance.